Abstract
Techniques are described for collaborative recommendation that constrain cross-user propagation of user-derived signals using asymmetric, type-dependent rules. A signal classification layer assigns each generated insight to a psychological signal type including positive preference, negative preference, behavioral pattern, and vulnerability indicator. An asymmetric propagation engine applies type-specific parameters such as hop depth, evidence thresholds, confidence discounting, and temporal decay, including higher friction for negative, behavioral, and vulnerability-related signals than for positive preferences. Negative preference propagation is constrained to a demotion-only effect in ranking, reducing scores without excluding candidates. A partitioned memory architecture stores ranking-accessible preferences in a first partition and stores behavioral patterns and vulnerability indicators in a second partition inaccessible to a ranking component. A well-being monitor reads both partitions and provides directives to the ranking component via a one-way constraint channel. Cascade monitoring may halt propagation based on growth and homogeneity measures.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Anonymous, "Systems and Methods for Asymmetric Propagation Constraints with Partitioned Memory Architecture in Collaborative Recommendation", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10757